xref: /freebsd/sys/vm/vm_mmap.c (revision daf1cffce2e07931f27c6c6998652e90df6ba87e)
1 /*
2  * Copyright (c) 1988 University of Utah.
3  * Copyright (c) 1991, 1993
4  *	The Regents of the University of California.  All rights reserved.
5  *
6  * This code is derived from software contributed to Berkeley by
7  * the Systems Programming Group of the University of Utah Computer
8  * Science Department.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  * from: Utah $Hdr: vm_mmap.c 1.6 91/10/21$
39  *
40  *	@(#)vm_mmap.c	8.4 (Berkeley) 1/12/94
41  * $FreeBSD$
42  */
43 
44 /*
45  * Mapped file (mmap) interface to VM
46  */
47 
48 #include "opt_compat.h"
49 #include "opt_rlimit.h"
50 
51 #include <sys/param.h>
52 #include <sys/kernel.h>
53 #include <sys/systm.h>
54 #include <sys/sysproto.h>
55 #include <sys/filedesc.h>
56 #include <sys/proc.h>
57 #include <sys/vnode.h>
58 #include <sys/fcntl.h>
59 #include <sys/file.h>
60 #include <sys/mman.h>
61 #include <sys/conf.h>
62 #include <sys/stat.h>
63 #include <sys/vmmeter.h>
64 #include <sys/sysctl.h>
65 
66 #include <vm/vm.h>
67 #include <vm/vm_param.h>
68 #include <sys/lock.h>
69 #include <vm/pmap.h>
70 #include <vm/vm_map.h>
71 #include <vm/vm_object.h>
72 #include <vm/vm_page.h>
73 #include <vm/vm_pager.h>
74 #include <vm/vm_pageout.h>
75 #include <vm/vm_extern.h>
76 #include <vm/vm_page.h>
77 #include <vm/vm_kern.h>
78 
79 #ifndef _SYS_SYSPROTO_H_
80 struct sbrk_args {
81 	int incr;
82 };
83 #endif
84 
85 static int max_proc_mmap;
86 SYSCTL_INT(_vm, OID_AUTO, max_proc_mmap, CTLFLAG_RW, &max_proc_mmap, 0, "");
87 
88 /*
89  * Set the maximum number of vm_map_entry structures per process.  Roughly
90  * speaking vm_map_entry structures are tiny, so allowing them to eat 1/100
91  * of our KVM malloc space still results in generous limits.  We want a
92  * default that is good enough to prevent the kernel running out of resources
93  * if attacked from compromised user account but generous enough such that
94  * multi-threaded processes are not unduly inconvenienced.
95  */
96 
97 static void vmmapentry_rsrc_init __P((void *));
98 SYSINIT(vmmersrc, SI_SUB_KVM_RSRC, SI_ORDER_FIRST, vmmapentry_rsrc_init, NULL)
99 
100 static void
101 vmmapentry_rsrc_init(dummy)
102         void *dummy;
103 {
104     max_proc_mmap = vm_kmem_size / sizeof(struct vm_map_entry);
105     max_proc_mmap /= 100;
106 }
107 
108 /* ARGSUSED */
109 int
110 sbrk(p, uap)
111 	struct proc *p;
112 	struct sbrk_args *uap;
113 {
114 
115 	/* Not yet implemented */
116 	return (EOPNOTSUPP);
117 }
118 
119 #ifndef _SYS_SYSPROTO_H_
120 struct sstk_args {
121 	int incr;
122 };
123 #endif
124 
125 /* ARGSUSED */
126 int
127 sstk(p, uap)
128 	struct proc *p;
129 	struct sstk_args *uap;
130 {
131 
132 	/* Not yet implemented */
133 	return (EOPNOTSUPP);
134 }
135 
136 #if defined(COMPAT_43) || defined(COMPAT_SUNOS)
137 #ifndef _SYS_SYSPROTO_H_
138 struct getpagesize_args {
139 	int dummy;
140 };
141 #endif
142 
143 /* ARGSUSED */
144 int
145 ogetpagesize(p, uap)
146 	struct proc *p;
147 	struct getpagesize_args *uap;
148 {
149 
150 	p->p_retval[0] = PAGE_SIZE;
151 	return (0);
152 }
153 #endif				/* COMPAT_43 || COMPAT_SUNOS */
154 
155 
156 /*
157  * Memory Map (mmap) system call.  Note that the file offset
158  * and address are allowed to be NOT page aligned, though if
159  * the MAP_FIXED flag it set, both must have the same remainder
160  * modulo the PAGE_SIZE (POSIX 1003.1b).  If the address is not
161  * page-aligned, the actual mapping starts at trunc_page(addr)
162  * and the return value is adjusted up by the page offset.
163  *
164  * Generally speaking, only character devices which are themselves
165  * memory-based, such as a video framebuffer, can be mmap'd.  Otherwise
166  * there would be no cache coherency between a descriptor and a VM mapping
167  * both to the same character device.
168  *
169  * Block devices can be mmap'd no matter what they represent.  Cache coherency
170  * is maintained as long as you do not write directly to the underlying
171  * character device.
172  */
173 #ifndef _SYS_SYSPROTO_H_
174 struct mmap_args {
175 	void *addr;
176 	size_t len;
177 	int prot;
178 	int flags;
179 	int fd;
180 	long pad;
181 	off_t pos;
182 };
183 #endif
184 
185 int
186 mmap(p, uap)
187 	struct proc *p;
188 	register struct mmap_args *uap;
189 {
190 	register struct filedesc *fdp = p->p_fd;
191 	register struct file *fp;
192 	struct vnode *vp;
193 	vm_offset_t addr;
194 	vm_size_t size, pageoff;
195 	vm_prot_t prot, maxprot;
196 	void *handle;
197 	int flags, error;
198 	int disablexworkaround;
199 	off_t pos;
200 	struct vmspace *vms = p->p_vmspace;
201 
202 	addr = (vm_offset_t) uap->addr;
203 	size = uap->len;
204 	prot = uap->prot & VM_PROT_ALL;
205 	flags = uap->flags;
206 	pos = uap->pos;
207 
208 	/* make sure mapping fits into numeric range etc */
209 	if ((ssize_t) uap->len < 0 ||
210 	    ((flags & MAP_ANON) && uap->fd != -1))
211 		return (EINVAL);
212 
213 	if (flags & MAP_STACK) {
214 		if ((uap->fd != -1) ||
215 		    ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
216 			return (EINVAL);
217 		flags |= MAP_ANON;
218 		pos = 0;
219 	}
220 
221 	/*
222 	 * Align the file position to a page boundary,
223 	 * and save its page offset component.
224 	 */
225 	pageoff = (pos & PAGE_MASK);
226 	pos -= pageoff;
227 
228 	/* Adjust size for rounding (on both ends). */
229 	size += pageoff;			/* low end... */
230 	size = (vm_size_t) round_page(size);	/* hi end */
231 
232 	/*
233 	 * Check for illegal addresses.  Watch out for address wrap... Note
234 	 * that VM_*_ADDRESS are not constants due to casts (argh).
235 	 */
236 	if (flags & MAP_FIXED) {
237 		/*
238 		 * The specified address must have the same remainder
239 		 * as the file offset taken modulo PAGE_SIZE, so it
240 		 * should be aligned after adjustment by pageoff.
241 		 */
242 		addr -= pageoff;
243 		if (addr & PAGE_MASK)
244 			return (EINVAL);
245 		/* Address range must be all in user VM space. */
246 		if (VM_MAXUSER_ADDRESS > 0 && addr + size > VM_MAXUSER_ADDRESS)
247 			return (EINVAL);
248 #ifndef i386
249 		if (VM_MIN_ADDRESS > 0 && addr < VM_MIN_ADDRESS)
250 			return (EINVAL);
251 #endif
252 		if (addr + size < addr)
253 			return (EINVAL);
254 	}
255 	/*
256 	 * XXX for non-fixed mappings where no hint is provided or
257 	 * the hint would fall in the potential heap space,
258 	 * place it after the end of the largest possible heap.
259 	 *
260 	 * There should really be a pmap call to determine a reasonable
261 	 * location.
262 	 */
263 	else if (addr == 0 ||
264 	    (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
265 	     addr < round_page((vm_offset_t)vms->vm_daddr + MAXDSIZ)))
266 		addr = round_page((vm_offset_t)vms->vm_daddr + MAXDSIZ);
267 
268 	if (flags & MAP_ANON) {
269 		/*
270 		 * Mapping blank space is trivial.
271 		 */
272 		handle = NULL;
273 		maxprot = VM_PROT_ALL;
274 		pos = 0;
275 	} else {
276 		/*
277 		 * Mapping file, get fp for validation. Obtain vnode and make
278 		 * sure it is of appropriate type.
279 		 */
280 		if (((unsigned) uap->fd) >= fdp->fd_nfiles ||
281 		    (fp = fdp->fd_ofiles[uap->fd]) == NULL)
282 			return (EBADF);
283 		if (fp->f_type != DTYPE_VNODE)
284 			return (EINVAL);
285 		vp = (struct vnode *) fp->f_data;
286 		if (vp->v_type != VREG && vp->v_type != VCHR)
287 			return (EINVAL);
288 		/*
289 		 * XXX hack to handle use of /dev/zero to map anon memory (ala
290 		 * SunOS).
291 		 */
292 		if (vp->v_type == VCHR && iszerodev(vp->v_rdev)) {
293 			handle = NULL;
294 			maxprot = VM_PROT_ALL;
295 			flags |= MAP_ANON;
296 			pos = 0;
297 		} else {
298 			/*
299 			 * cdevs does not provide private mappings of any kind.
300 			 */
301 			/*
302 			 * However, for XIG X server to continue to work,
303 			 * we should allow the superuser to do it anyway.
304 			 * We only allow it at securelevel < 1.
305 			 * (Because the XIG X server writes directly to video
306 			 * memory via /dev/mem, it should never work at any
307 			 * other securelevel.
308 			 * XXX this will have to go
309 			 */
310 			if (securelevel >= 1)
311 				disablexworkaround = 1;
312 			else
313 				disablexworkaround = suser(p);
314 			if (vp->v_type == VCHR && disablexworkaround &&
315 				(flags & (MAP_PRIVATE|MAP_COPY)))
316 				 return (EINVAL);
317 			/*
318 			 * Ensure that file and memory protections are
319 			 * compatible.  Note that we only worry about
320 			 * writability if mapping is shared; in this case,
321 			 * current and max prot are dictated by the open file.
322 			 * XXX use the vnode instead?  Problem is: what
323 			 * credentials do we use for determination? What if
324 			 * proc does a setuid?
325 			 */
326 			maxprot = VM_PROT_EXECUTE;	/* ??? */
327 			if (fp->f_flag & FREAD)
328 				maxprot |= VM_PROT_READ;
329 			else if (prot & PROT_READ)
330 				return (EACCES);
331 			/*
332 			 * If we are sharing potential changes (either via
333 			 * MAP_SHARED or via the implicit sharing of character
334 			 * device mappings), and we are trying to get write
335 			 * permission although we opened it without asking
336 			 * for it, bail out.  Check for superuser, only if
337 			 * we're at securelevel < 1, to allow the XIG X server
338 			 * to continue to work.
339 			 */
340 
341 			if ((flags & MAP_SHARED) != 0 ||
342 			    (vp->v_type == VCHR && disablexworkaround)) {
343 				if ((fp->f_flag & FWRITE) != 0) {
344 					struct vattr va;
345 					if ((error =
346 					    VOP_GETATTR(vp, &va,
347 						        p->p_ucred, p)))
348 						return (error);
349 					if ((va.va_flags &
350 					    (IMMUTABLE|APPEND)) == 0)
351 						maxprot |= VM_PROT_WRITE;
352 					else if (prot & PROT_WRITE)
353 						return (EPERM);
354 				} else if ((prot & PROT_WRITE) != 0)
355 					return (EACCES);
356 			} else
357 				maxprot |= VM_PROT_WRITE;
358 
359 			handle = (void *)vp;
360 		}
361 	}
362 
363 	/*
364 	 * Do not allow more then a certain number of vm_map_entry structures
365 	 * per process.  Scale with the number of rforks sharing the map
366 	 * to make the limit reasonable for threads.
367 	 */
368 	if (max_proc_mmap &&
369 	    vms->vm_map.nentries >= max_proc_mmap * vms->vm_refcnt) {
370 		return (ENOMEM);
371 	}
372 
373 	error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
374 	    flags, handle, pos);
375 	if (error == 0)
376 		p->p_retval[0] = (register_t) (addr + pageoff);
377 	return (error);
378 }
379 
380 #ifdef COMPAT_43
381 #ifndef _SYS_SYSPROTO_H_
382 struct ommap_args {
383 	caddr_t addr;
384 	int len;
385 	int prot;
386 	int flags;
387 	int fd;
388 	long pos;
389 };
390 #endif
391 int
392 ommap(p, uap)
393 	struct proc *p;
394 	register struct ommap_args *uap;
395 {
396 	struct mmap_args nargs;
397 	static const char cvtbsdprot[8] = {
398 		0,
399 		PROT_EXEC,
400 		PROT_WRITE,
401 		PROT_EXEC | PROT_WRITE,
402 		PROT_READ,
403 		PROT_EXEC | PROT_READ,
404 		PROT_WRITE | PROT_READ,
405 		PROT_EXEC | PROT_WRITE | PROT_READ,
406 	};
407 
408 #define	OMAP_ANON	0x0002
409 #define	OMAP_COPY	0x0020
410 #define	OMAP_SHARED	0x0010
411 #define	OMAP_FIXED	0x0100
412 #define	OMAP_INHERIT	0x0800
413 
414 	nargs.addr = uap->addr;
415 	nargs.len = uap->len;
416 	nargs.prot = cvtbsdprot[uap->prot & 0x7];
417 	nargs.flags = 0;
418 	if (uap->flags & OMAP_ANON)
419 		nargs.flags |= MAP_ANON;
420 	if (uap->flags & OMAP_COPY)
421 		nargs.flags |= MAP_COPY;
422 	if (uap->flags & OMAP_SHARED)
423 		nargs.flags |= MAP_SHARED;
424 	else
425 		nargs.flags |= MAP_PRIVATE;
426 	if (uap->flags & OMAP_FIXED)
427 		nargs.flags |= MAP_FIXED;
428 	if (uap->flags & OMAP_INHERIT)
429 		nargs.flags |= MAP_INHERIT;
430 	nargs.fd = uap->fd;
431 	nargs.pos = uap->pos;
432 	return (mmap(p, &nargs));
433 }
434 #endif				/* COMPAT_43 */
435 
436 
437 #ifndef _SYS_SYSPROTO_H_
438 struct msync_args {
439 	void *addr;
440 	int len;
441 	int flags;
442 };
443 #endif
444 int
445 msync(p, uap)
446 	struct proc *p;
447 	struct msync_args *uap;
448 {
449 	vm_offset_t addr;
450 	vm_size_t size, pageoff;
451 	int flags;
452 	vm_map_t map;
453 	int rv;
454 
455 	addr = (vm_offset_t) uap->addr;
456 	size = uap->len;
457 	flags = uap->flags;
458 
459 	pageoff = (addr & PAGE_MASK);
460 	addr -= pageoff;
461 	size += pageoff;
462 	size = (vm_size_t) round_page(size);
463 	if (addr + size < addr)
464 		return(EINVAL);
465 
466 	if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
467 		return (EINVAL);
468 
469 	map = &p->p_vmspace->vm_map;
470 
471 	/*
472 	 * XXX Gak!  If size is zero we are supposed to sync "all modified
473 	 * pages with the region containing addr".  Unfortunately, we don't
474 	 * really keep track of individual mmaps so we approximate by flushing
475 	 * the range of the map entry containing addr. This can be incorrect
476 	 * if the region splits or is coalesced with a neighbor.
477 	 */
478 	if (size == 0) {
479 		vm_map_entry_t entry;
480 
481 		vm_map_lock_read(map);
482 		rv = vm_map_lookup_entry(map, addr, &entry);
483 		vm_map_unlock_read(map);
484 		if (rv == FALSE)
485 			return (EINVAL);
486 		addr = entry->start;
487 		size = entry->end - entry->start;
488 	}
489 
490 	/*
491 	 * Clean the pages and interpret the return value.
492 	 */
493 	rv = vm_map_clean(map, addr, addr + size, (flags & MS_ASYNC) == 0,
494 	    (flags & MS_INVALIDATE) != 0);
495 
496 	switch (rv) {
497 	case KERN_SUCCESS:
498 		break;
499 	case KERN_INVALID_ADDRESS:
500 		return (EINVAL);	/* Sun returns ENOMEM? */
501 	case KERN_FAILURE:
502 		return (EIO);
503 	default:
504 		return (EINVAL);
505 	}
506 
507 	return (0);
508 }
509 
510 #ifndef _SYS_SYSPROTO_H_
511 struct munmap_args {
512 	void *addr;
513 	size_t len;
514 };
515 #endif
516 int
517 munmap(p, uap)
518 	register struct proc *p;
519 	register struct munmap_args *uap;
520 {
521 	vm_offset_t addr;
522 	vm_size_t size, pageoff;
523 	vm_map_t map;
524 
525 	addr = (vm_offset_t) uap->addr;
526 	size = uap->len;
527 
528 	pageoff = (addr & PAGE_MASK);
529 	addr -= pageoff;
530 	size += pageoff;
531 	size = (vm_size_t) round_page(size);
532 	if (addr + size < addr)
533 		return(EINVAL);
534 
535 	if (size == 0)
536 		return (0);
537 
538 	/*
539 	 * Check for illegal addresses.  Watch out for address wrap... Note
540 	 * that VM_*_ADDRESS are not constants due to casts (argh).
541 	 */
542 	if (VM_MAXUSER_ADDRESS > 0 && addr + size > VM_MAXUSER_ADDRESS)
543 		return (EINVAL);
544 #ifndef i386
545 	if (VM_MIN_ADDRESS > 0 && addr < VM_MIN_ADDRESS)
546 		return (EINVAL);
547 #endif
548 	map = &p->p_vmspace->vm_map;
549 	/*
550 	 * Make sure entire range is allocated.
551 	 */
552 	if (!vm_map_check_protection(map, addr, addr + size, VM_PROT_NONE))
553 		return (EINVAL);
554 	/* returns nothing but KERN_SUCCESS anyway */
555 	(void) vm_map_remove(map, addr, addr + size);
556 	return (0);
557 }
558 
559 void
560 munmapfd(p, fd)
561 	struct proc *p;
562 	int fd;
563 {
564 	/*
565 	 * XXX should unmap any regions mapped to this file
566 	 */
567 	p->p_fd->fd_ofileflags[fd] &= ~UF_MAPPED;
568 }
569 
570 #ifndef _SYS_SYSPROTO_H_
571 struct mprotect_args {
572 	const void *addr;
573 	size_t len;
574 	int prot;
575 };
576 #endif
577 int
578 mprotect(p, uap)
579 	struct proc *p;
580 	struct mprotect_args *uap;
581 {
582 	vm_offset_t addr;
583 	vm_size_t size, pageoff;
584 	register vm_prot_t prot;
585 
586 	addr = (vm_offset_t) uap->addr;
587 	size = uap->len;
588 	prot = uap->prot & VM_PROT_ALL;
589 #if defined(VM_PROT_READ_IS_EXEC)
590 	if (prot & VM_PROT_READ)
591 		prot |= VM_PROT_EXECUTE;
592 #endif
593 
594 	pageoff = (addr & PAGE_MASK);
595 	addr -= pageoff;
596 	size += pageoff;
597 	size = (vm_size_t) round_page(size);
598 	if (addr + size < addr)
599 		return(EINVAL);
600 
601 	switch (vm_map_protect(&p->p_vmspace->vm_map, addr, addr + size, prot,
602 		FALSE)) {
603 	case KERN_SUCCESS:
604 		return (0);
605 	case KERN_PROTECTION_FAILURE:
606 		return (EACCES);
607 	}
608 	return (EINVAL);
609 }
610 
611 #ifndef _SYS_SYSPROTO_H_
612 struct minherit_args {
613 	void *addr;
614 	size_t len;
615 	int inherit;
616 };
617 #endif
618 int
619 minherit(p, uap)
620 	struct proc *p;
621 	struct minherit_args *uap;
622 {
623 	vm_offset_t addr;
624 	vm_size_t size, pageoff;
625 	register vm_inherit_t inherit;
626 
627 	addr = (vm_offset_t)uap->addr;
628 	size = uap->len;
629 	inherit = uap->inherit;
630 
631 	pageoff = (addr & PAGE_MASK);
632 	addr -= pageoff;
633 	size += pageoff;
634 	size = (vm_size_t) round_page(size);
635 	if (addr + size < addr)
636 		return(EINVAL);
637 
638 	switch (vm_map_inherit(&p->p_vmspace->vm_map, addr, addr+size,
639 	    inherit)) {
640 	case KERN_SUCCESS:
641 		return (0);
642 	case KERN_PROTECTION_FAILURE:
643 		return (EACCES);
644 	}
645 	return (EINVAL);
646 }
647 
648 #ifndef _SYS_SYSPROTO_H_
649 struct madvise_args {
650 	void *addr;
651 	size_t len;
652 	int behav;
653 };
654 #endif
655 
656 /* ARGSUSED */
657 int
658 madvise(p, uap)
659 	struct proc *p;
660 	struct madvise_args *uap;
661 {
662 	vm_offset_t start, end;
663 
664 	/*
665 	 * Check for illegal behavior
666 	 */
667 	if (uap->behav < 0 || uap->behav > MADV_CORE)
668 		return (EINVAL);
669 	/*
670 	 * Check for illegal addresses.  Watch out for address wrap... Note
671 	 * that VM_*_ADDRESS are not constants due to casts (argh).
672 	 */
673 	if (VM_MAXUSER_ADDRESS > 0 &&
674 		((vm_offset_t) uap->addr + uap->len) > VM_MAXUSER_ADDRESS)
675 		return (EINVAL);
676 #ifndef i386
677 	if (VM_MIN_ADDRESS > 0 && uap->addr < VM_MIN_ADDRESS)
678 		return (EINVAL);
679 #endif
680 	if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr)
681 		return (EINVAL);
682 
683 	/*
684 	 * Since this routine is only advisory, we default to conservative
685 	 * behavior.
686 	 */
687 	start = trunc_page((vm_offset_t) uap->addr);
688 	end = round_page((vm_offset_t) uap->addr + uap->len);
689 
690 	if (vm_map_madvise(&p->p_vmspace->vm_map, start, end, uap->behav))
691 		return (EINVAL);
692 	return (0);
693 }
694 
695 #ifndef _SYS_SYSPROTO_H_
696 struct mincore_args {
697 	const void *addr;
698 	size_t len;
699 	char *vec;
700 };
701 #endif
702 
703 /* ARGSUSED */
704 int
705 mincore(p, uap)
706 	struct proc *p;
707 	struct mincore_args *uap;
708 {
709 	vm_offset_t addr, first_addr;
710 	vm_offset_t end, cend;
711 	pmap_t pmap;
712 	vm_map_t map;
713 	char *vec;
714 	int error;
715 	int vecindex, lastvecindex;
716 	register vm_map_entry_t current;
717 	vm_map_entry_t entry;
718 	int mincoreinfo;
719 	unsigned int timestamp;
720 
721 	/*
722 	 * Make sure that the addresses presented are valid for user
723 	 * mode.
724 	 */
725 	first_addr = addr = trunc_page((vm_offset_t) uap->addr);
726 	end = addr + (vm_size_t)round_page(uap->len);
727 	if (VM_MAXUSER_ADDRESS > 0 && end > VM_MAXUSER_ADDRESS)
728 		return (EINVAL);
729 	if (end < addr)
730 		return (EINVAL);
731 
732 	/*
733 	 * Address of byte vector
734 	 */
735 	vec = uap->vec;
736 
737 	map = &p->p_vmspace->vm_map;
738 	pmap = vmspace_pmap(p->p_vmspace);
739 
740 	vm_map_lock_read(map);
741 RestartScan:
742 	timestamp = map->timestamp;
743 
744 	if (!vm_map_lookup_entry(map, addr, &entry))
745 		entry = entry->next;
746 
747 	/*
748 	 * Do this on a map entry basis so that if the pages are not
749 	 * in the current processes address space, we can easily look
750 	 * up the pages elsewhere.
751 	 */
752 	lastvecindex = -1;
753 	for(current = entry;
754 		(current != &map->header) && (current->start < end);
755 		current = current->next) {
756 
757 		/*
758 		 * ignore submaps (for now) or null objects
759 		 */
760 		if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
761 			current->object.vm_object == NULL)
762 			continue;
763 
764 		/*
765 		 * limit this scan to the current map entry and the
766 		 * limits for the mincore call
767 		 */
768 		if (addr < current->start)
769 			addr = current->start;
770 		cend = current->end;
771 		if (cend > end)
772 			cend = end;
773 
774 		/*
775 		 * scan this entry one page at a time
776 		 */
777 		while(addr < cend) {
778 			/*
779 			 * Check pmap first, it is likely faster, also
780 			 * it can provide info as to whether we are the
781 			 * one referencing or modifying the page.
782 			 */
783 			mincoreinfo = pmap_mincore(pmap, addr);
784 			if (!mincoreinfo) {
785 				vm_pindex_t pindex;
786 				vm_ooffset_t offset;
787 				vm_page_t m;
788 				/*
789 				 * calculate the page index into the object
790 				 */
791 				offset = current->offset + (addr - current->start);
792 				pindex = OFF_TO_IDX(offset);
793 				m = vm_page_lookup(current->object.vm_object,
794 					pindex);
795 				/*
796 				 * if the page is resident, then gather information about
797 				 * it.
798 				 */
799 				if (m) {
800 					mincoreinfo = MINCORE_INCORE;
801 					if (m->dirty ||
802 						pmap_is_modified(VM_PAGE_TO_PHYS(m)))
803 						mincoreinfo |= MINCORE_MODIFIED_OTHER;
804 					if ((m->flags & PG_REFERENCED) ||
805 						pmap_ts_referenced(VM_PAGE_TO_PHYS(m))) {
806 						vm_page_flag_set(m, PG_REFERENCED);
807 						mincoreinfo |= MINCORE_REFERENCED_OTHER;
808 					}
809 				}
810 			}
811 
812 			/*
813 			 * subyte may page fault.  In case it needs to modify
814 			 * the map, we release the lock.
815 			 */
816 			vm_map_unlock_read(map);
817 
818 			/*
819 			 * calculate index into user supplied byte vector
820 			 */
821 			vecindex = OFF_TO_IDX(addr - first_addr);
822 
823 			/*
824 			 * If we have skipped map entries, we need to make sure that
825 			 * the byte vector is zeroed for those skipped entries.
826 			 */
827 			while((lastvecindex + 1) < vecindex) {
828 				error = subyte( vec + lastvecindex, 0);
829 				if (error) {
830 					return (EFAULT);
831 				}
832 				++lastvecindex;
833 			}
834 
835 			/*
836 			 * Pass the page information to the user
837 			 */
838 			error = subyte( vec + vecindex, mincoreinfo);
839 			if (error) {
840 				return (EFAULT);
841 			}
842 
843 			/*
844 			 * If the map has changed, due to the subyte, the previous
845 			 * output may be invalid.
846 			 */
847 			vm_map_lock_read(map);
848 			if (timestamp != map->timestamp)
849 				goto RestartScan;
850 
851 			lastvecindex = vecindex;
852 			addr += PAGE_SIZE;
853 		}
854 	}
855 
856 	/*
857 	 * subyte may page fault.  In case it needs to modify
858 	 * the map, we release the lock.
859 	 */
860 	vm_map_unlock_read(map);
861 
862 	/*
863 	 * Zero the last entries in the byte vector.
864 	 */
865 	vecindex = OFF_TO_IDX(end - first_addr);
866 	while((lastvecindex + 1) < vecindex) {
867 		error = subyte( vec + lastvecindex, 0);
868 		if (error) {
869 			return (EFAULT);
870 		}
871 		++lastvecindex;
872 	}
873 
874 	/*
875 	 * If the map has changed, due to the subyte, the previous
876 	 * output may be invalid.
877 	 */
878 	vm_map_lock_read(map);
879 	if (timestamp != map->timestamp)
880 		goto RestartScan;
881 	vm_map_unlock_read(map);
882 
883 	return (0);
884 }
885 
886 #ifndef _SYS_SYSPROTO_H_
887 struct mlock_args {
888 	const void *addr;
889 	size_t len;
890 };
891 #endif
892 int
893 mlock(p, uap)
894 	struct proc *p;
895 	struct mlock_args *uap;
896 {
897 	vm_offset_t addr;
898 	vm_size_t size, pageoff;
899 	int error;
900 
901 	addr = (vm_offset_t) uap->addr;
902 	size = uap->len;
903 
904 	pageoff = (addr & PAGE_MASK);
905 	addr -= pageoff;
906 	size += pageoff;
907 	size = (vm_size_t) round_page(size);
908 
909 	/* disable wrap around */
910 	if (addr + size < addr)
911 		return (EINVAL);
912 
913 	if (atop(size) + cnt.v_wire_count > vm_page_max_wired)
914 		return (EAGAIN);
915 
916 #ifdef pmap_wired_count
917 	if (size + ptoa(pmap_wired_count(vm_map_pmap(&p->p_vmspace->vm_map))) >
918 	    p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur)
919 		return (ENOMEM);
920 #else
921 	error = suser(p);
922 	if (error)
923 		return (error);
924 #endif
925 
926 	error = vm_map_user_pageable(&p->p_vmspace->vm_map, addr, addr + size, FALSE);
927 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
928 }
929 
930 #ifndef _SYS_SYSPROTO_H_
931 struct mlockall_args {
932 	int	how;
933 };
934 #endif
935 
936 int
937 mlockall(p, uap)
938 	struct proc *p;
939 	struct mlockall_args *uap;
940 {
941 	return 0;
942 }
943 
944 #ifndef _SYS_SYSPROTO_H_
945 struct mlockall_args {
946 	int	how;
947 };
948 #endif
949 
950 int
951 munlockall(p, uap)
952 	struct proc *p;
953 	struct munlockall_args *uap;
954 {
955 	return 0;
956 }
957 
958 #ifndef _SYS_SYSPROTO_H_
959 struct munlock_args {
960 	const void *addr;
961 	size_t len;
962 };
963 #endif
964 int
965 munlock(p, uap)
966 	struct proc *p;
967 	struct munlock_args *uap;
968 {
969 	vm_offset_t addr;
970 	vm_size_t size, pageoff;
971 	int error;
972 
973 	addr = (vm_offset_t) uap->addr;
974 	size = uap->len;
975 
976 	pageoff = (addr & PAGE_MASK);
977 	addr -= pageoff;
978 	size += pageoff;
979 	size = (vm_size_t) round_page(size);
980 
981 	/* disable wrap around */
982 	if (addr + size < addr)
983 		return (EINVAL);
984 
985 #ifndef pmap_wired_count
986 	error = suser(p);
987 	if (error)
988 		return (error);
989 #endif
990 
991 	error = vm_map_user_pageable(&p->p_vmspace->vm_map, addr, addr + size, TRUE);
992 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
993 }
994 
995 /*
996  * Internal version of mmap.
997  * Currently used by mmap, exec, and sys5 shared memory.
998  * Handle is either a vnode pointer or NULL for MAP_ANON.
999  */
1000 int
1001 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1002 	vm_prot_t maxprot, int flags,
1003 	void *handle,
1004 	vm_ooffset_t foff)
1005 {
1006 	boolean_t fitit;
1007 	vm_object_t object;
1008 	struct vnode *vp = NULL;
1009 	objtype_t type;
1010 	int rv = KERN_SUCCESS;
1011 	vm_ooffset_t objsize;
1012 	int docow;
1013 	struct proc *p = curproc;
1014 
1015 	if (size == 0)
1016 		return (0);
1017 
1018 	objsize = size = round_page(size);
1019 
1020 	/*
1021 	 * We currently can only deal with page aligned file offsets.
1022 	 * The check is here rather than in the syscall because the
1023 	 * kernel calls this function internally for other mmaping
1024 	 * operations (such as in exec) and non-aligned offsets will
1025 	 * cause pmap inconsistencies...so we want to be sure to
1026 	 * disallow this in all cases.
1027 	 */
1028 	if (foff & PAGE_MASK)
1029 		return (EINVAL);
1030 
1031 	if ((flags & MAP_FIXED) == 0) {
1032 		fitit = TRUE;
1033 		*addr = round_page(*addr);
1034 	} else {
1035 		if (*addr != trunc_page(*addr))
1036 			return (EINVAL);
1037 		fitit = FALSE;
1038 		(void) vm_map_remove(map, *addr, *addr + size);
1039 	}
1040 
1041 	/*
1042 	 * Lookup/allocate object.
1043 	 */
1044 	if (flags & MAP_ANON) {
1045 		type = OBJT_DEFAULT;
1046 		/*
1047 		 * Unnamed anonymous regions always start at 0.
1048 		 */
1049 		if (handle == 0)
1050 			foff = 0;
1051 	} else {
1052 		vp = (struct vnode *) handle;
1053 		if (vp->v_type == VCHR) {
1054 			type = OBJT_DEVICE;
1055 			handle = (void *)(intptr_t)vp->v_rdev;
1056 		} else {
1057 			struct vattr vat;
1058 			int error;
1059 
1060 			error = VOP_GETATTR(vp, &vat, p->p_ucred, p);
1061 			if (error)
1062 				return (error);
1063 			objsize = round_page(vat.va_size);
1064 			type = OBJT_VNODE;
1065 			/*
1066 			 * if it is a regular file without any references
1067 			 * we do not need to sync it.
1068 			 */
1069 			if (vp->v_type == VREG && vat.va_nlink == 0) {
1070 				flags |= MAP_NOSYNC;
1071 			}
1072 		}
1073 	}
1074 
1075 	if (handle == NULL) {
1076 		object = NULL;
1077 		docow = 0;
1078 	} else {
1079 		object = vm_pager_allocate(type,
1080 			handle, objsize, prot, foff);
1081 		if (object == NULL)
1082 			return (type == OBJT_DEVICE ? EINVAL : ENOMEM);
1083 		docow = MAP_PREFAULT_PARTIAL;
1084 	}
1085 
1086 	/*
1087 	 * Force device mappings to be shared.
1088 	 */
1089 	if (type == OBJT_DEVICE) {
1090 		flags &= ~(MAP_PRIVATE|MAP_COPY);
1091 		flags |= MAP_SHARED;
1092 	}
1093 
1094 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1095 		docow |= MAP_COPY_ON_WRITE;
1096 	if (flags & MAP_NOSYNC)
1097 		docow |= MAP_DISABLE_SYNCER;
1098 	if (flags & MAP_NOCORE)
1099 		docow |= MAP_DISABLE_COREDUMP;
1100 
1101 #if defined(VM_PROT_READ_IS_EXEC)
1102 	if (prot & VM_PROT_READ)
1103 		prot |= VM_PROT_EXECUTE;
1104 
1105 	if (maxprot & VM_PROT_READ)
1106 		maxprot |= VM_PROT_EXECUTE;
1107 #endif
1108 
1109 	if (fitit) {
1110 		*addr = pmap_addr_hint(object, *addr, size);
1111 	}
1112 
1113 	if (flags & MAP_STACK)
1114 		rv = vm_map_stack (map, *addr, size, prot,
1115 				   maxprot, docow);
1116 	else
1117 		rv = vm_map_find(map, object, foff, addr, size, fitit,
1118 				 prot, maxprot, docow);
1119 
1120 	if (rv != KERN_SUCCESS) {
1121 		/*
1122 		 * Lose the object reference. Will destroy the
1123 		 * object if it's an unnamed anonymous mapping
1124 		 * or named anonymous without other references.
1125 		 */
1126 		vm_object_deallocate(object);
1127 		goto out;
1128 	}
1129 
1130 	/*
1131 	 * Shared memory is also shared with children.
1132 	 */
1133 	if (flags & (MAP_SHARED|MAP_INHERIT)) {
1134 		rv = vm_map_inherit(map, *addr, *addr + size, VM_INHERIT_SHARE);
1135 		if (rv != KERN_SUCCESS) {
1136 			(void) vm_map_remove(map, *addr, *addr + size);
1137 			goto out;
1138 		}
1139 	}
1140 out:
1141 	switch (rv) {
1142 	case KERN_SUCCESS:
1143 		return (0);
1144 	case KERN_INVALID_ADDRESS:
1145 	case KERN_NO_SPACE:
1146 		return (ENOMEM);
1147 	case KERN_PROTECTION_FAILURE:
1148 		return (EACCES);
1149 	default:
1150 		return (EINVAL);
1151 	}
1152 }
1153